The chip shortage never really ended

Remember 2021? You couldn’t buy a PS5. Car dealerships had empty lots. People were paying double for graphics cards that weren’t even good. The chip shortage was everywhere, all at once, and everyone noticed.

Then around mid-2023 the headlines stopped. “Chip shortage easing.” “Supply chains recovering.” “Crisis over.” And consumer electronics got easier to buy again, so most people moved on.

But the shortage didn’t end. It shapeshifted.

What happened is that the consumer chip shortage became an AI chip shortage. NVIDIA went from a gaming company to probably the most important company in the world, and suddenly the bottleneck wasn’t PS5 processors. It was H100s. Then B200s. Then whatever comes next.

TSMC is running their advanced nodes at full capacity. They’ve been doing that since 2022 and they haven’t stopped. The Arizona fab is producing chips now but it barely dents the demand. Samsung’s 2nm process is behind schedule. Intel’s foundry ambitions are… let’s call them “evolving.”

The thing people don’t talk about enough is water. TSMC’s fabs in Hsinchu use something like 150,000 tons of water per day. Per day. Taiwan had a drought scare in 2021 that almost shut down production. The semiconductor industry’s biggest risk isn’t a competitor. It’s weather.

I was reading about ASML’s new High-NA EUV machines last week. Each one costs around $380 million. There are maybe a dozen of them in the world. The entire future of computing depends on a small number of machines built by one company in the Netherlands, installed in fabs that need more water than some cities, producing chips measured in atoms.

And we need more of them. A lot more. Because every AI model, every self-driving car, every robot, every smart anything needs chips. The demand curve points up and to the right with no sign of flattening.

So no. The chip shortage didn’t end. We just stopped calling it that because the people affected changed. It used to be gamers and car buyers. Now it’s AI labs and data centers. The fundamentals haven’t changed: we want more compute than we can physically manufacture.

I think about this a lot. We’re building an entire civilization layer on top of technology that depends on a few fabs, a few machines, a lot of water, and the assumption that nothing goes wrong in any of those supply chains simultaneously.

That’s a lot of assumptions.